China Financial Conditions Index

01

Two indices, one engine

Both indices use the same eight components, the same signs, and the same category weights. They differ in one thing only: whether each series is standardized against a fixed historical base or a rolling window.

02

Components & weights

Eight series in five categories. The sign encodes a growth-impulse convention: +1 means a higher value tightens conditions (raises the index); −1 means a higher value loosens them (lowers it).

Category weights sum to 1.00; sub-weights sum to 1.00 within each category. Equal-weight (0.20 each) is the validation benchmark; FX is cut to 0.10 (a managed float carries little information) and the freed weight routed to the credit impulse.

03

How it's built

1 · Sign & standardize

Each raw series is signed and standardized to zero mean, unit variance:

x_i,t = sign_i × ( X_i,t − μ_i ) / σ_i

The moments (μ, σ) are the only switch between the two indices — the full 2018-present sample for the Level, a trailing 52-week window for Momentum.

2 · Category composites

Multi-series categories (FX, Equity, Rates) are a plain weighted average of their component z-scores — with no second standardization. This keeps the index fully transparent: the Momentum ends up as exactly the weighted sum of the eight tile z-scores, so every number on the dashboard reconciles by hand (see §07).

fx     = 0.5·x_cfets  + 0.5·x_reer
equity = 0.5·x_gap    + 0.5·x_vol
rates  = 0.5·x_shibor + 0.5·x_cgb10y
property, credit  =  their single z-score (already one series)

Trade-off: averaging correlated sub-series (FX most of all) slightly shrinks that category's variance, so its effective influence runs a touch below its nominal weight. We accept that in exchange for an index you can verify by hand from the tiles.

3 · Weight & aggregate

C_t = 0.10·FX + 0.20·Equity + 0.20·Rates + 0.20·Property + 0.30·Credit

4 · Anchor

Level     :  FCI = 100 + (1 / SD_base(C)) × C_t
Momentum  :  FCI = C_t          (rolling moments)

This anchoring is our own scaling choice: because the deviation is divided by the composite's full-sample standard deviation, one index point equals one base-period SD by construction — so 98 is ~2 SD looser than the post-2018 average and 102 ~2 SD tighter. It is the same SD scale the commentary uses when it cites, e.g., "−0.1 SD". Note this is not how Goldman scales its headline index (see §06). Master frequency is weekly (Friday); daily series are sampled last-obs, monthly series (property, credit) forward-filled with no interpolation.

Because the Momentum is a weighted average of z-scores rather than a single standardized series, its absolute scale is compressed (it rarely reaches ±1). Its headline reading is therefore expressed as a percentile of its own history — the label ("tightening fast" etc.) keys off that percentile, not a fixed z-threshold.

04

Sign conventions growth-impulse

Signs follow the logic that asset-price and credit strength is stimulative — broadly the framing Goldman uses. This is a deliberate choice; it makes the index a read on the growth impulse from conditions, not a froth/stress gauge.

• Weak RMB → looser. A weaker trade-weighted RMB supports an export economy, so it eases conditions. (Strong RMB → tighter.) The RMB is a managed float, so it moves little and in administered steps — hence FX carries the lowest weight.
• Equities above trend → looser via the wealth effect and cheaper equity capital. China's household equity-wealth share is low (most wealth is in property), so equity is a smaller channel here than in the US.
• Equity volatility → tighter. Elevated vol signals stress — and deliberately offsets the trend gap (see §6).
• Higher SHIBOR / 10Y CGB → tighter (higher cost of capital). The short end (SHIBOR) is liquidity-driven; the long end (CGB) is the cost-of-capital anchor.
• Faster house-price growth → looser via collateral and wealth. The 70-city second-hand series is used rather than new-home prices, which are partly administered/floored — 2nd-hand is the cleaner market signal.
• Accelerating credit → looser. The TSF credit impulse is the single most important China financial-conditions variable; in China, policy transmits through the quantity of credit, not its price.

05

Reading the dashboard

• FCI — Level (top): the headline. Green below 100 = looser than the post-2018 norm, red above = tighter.
• Category contribution (directly beneath): decomposes the current Level into its five category contributions — green easing, red tightening. This explains the chart above it.
• FCI — Momentum: the rolling z-score. The shaded grey band marks the trailing 52-week window the current z is measured over.
• Component tiles: each raw series over its last 52 weeks, with a signed z-badge (green easing / red tightening) measured versus the last year. The SHCOMP tile shows its 52-week price gain.

06

Relationship to Goldman's FCI

The Level index is built on the methodology Goldman set out in Our New G10 Financial Conditions Indices (Goldman Sachs Global Economics, 2017): a weighted average of a short rate, a long-term yield, a credit measure, an equity-price variable and a trade-weighted exchange rate, with weights reflecting each variable's estimated effect on GDP growth over a one-year horizon.

This index departs from Goldman's daily China FCI in two deliberate ways. First, it adds a property channel and a TSF credit-impulse channel — together 50% of the weight — which the daily GS index omits. Second, the equity leg uses a trend gap plus a realised-vol overlay rather than a valuation (Shiller-style) input. The upshot: when the property/credit cycle is weak, this Level reads meaningfully less tight than a rates-dominated GSCNFCI, because half the index sits on the downturn channels that offset easy money.

On units. Our Level expresses deviations in post-2018 standard deviations (1 point = 1 SD; see §03). Goldman's headline index is instead scaled to growth impact — a one-point move corresponds to roughly one percentage point of year-ahead GDP growth — so its "points" are not standard deviations. Standardized, SD-based presentations of the GS FCI do exist (the BIS, for example, publishes it as 100 = long-run average with each unit a one-SD move), and our scale matches that convention rather than the paper's native growth-impact units.

07

Reconciling the Momentum from the components

The rolling Momentum is simply the weighted sum of the eight component z-scores on the dashboard tiles. Each tile shows both numbers you need: its z-score (the coloured badge, e.g. +1.8σ) and its weight (e.g. weight 30%). Multiply the two for every tile and add them up:

Momentum  =  Σ ( tile z-score × weight )
weight    =  sub-weight × category weight   (fixed; sums to 100%)

Worked on a recent week's tiles (substitute whatever your tiles currently read):

Same thing read by category (averaging is order-independent): FX +0.17, Equity +0.34, Rates −0.24, Property +0.02, Credit +0.54 → +0.83.

Two things this makes clear. First, a component's pull depends on its z-score and its weight together: credit at +1.8σ moves the index far more than CFETS at +2.3σ, because credit carries 30% versus CFETS's 5%. Second, the headline (+0.83 here) is small in absolute terms precisely because it is an average of z-scores — so its standing is read as a percentile of the Momentum's own history (§03), which is how a modest +0.83 can sit near the 90th percentile: unusually tight for this cycle.

08

Data & refresh

History is seeded once from a Bloomberg export; from then on every series updates from a free feed, so no weekly Bloomberg export is needed:

• SHCOMP — AkShare stock_zh_index_daily, daily
• 3M SHIBOR & CGB 10Y — AkShare rate_interbank and bond_zh_us_rate, daily
• REER — FRED RBCNBIS (BIS broad, no key), monthly
• CFETS RMB Index — CFETS via chinamoney.com.cn, daily
• 70-city 2nd-hand prices — NBS via the Eastmoney datacenter (mean of the 70 cities), monthly
• TSF credit impulse — AkShare monthly TSF flow, cumulated onto the seeded stock and 12-month differenced, monthly

Published weekly, Friday morning (06:00 SGT), to Discord with an auto-generated commentary.

09

Caveats

• Short sample (2018→). Bloomberg's TSF stock series begins Jan-2017 (a PBoC rebasing); the 12-month credit impulse off it is first valid Jan-2018 and binds the whole common sample. A frozen pre-period "normal" like the Korea build isn't available — the Level uses a full-sample base instead. Swapping the credit channel to M2 YoY would extend history to ~2009 at the cost of a narrower credit definition.
• Credit impulse is a second derivative — the change of a growth rate, so inherently noisy. At 0.30 weight it can whip Momentum around; winsorising it or dialling the weight to 0.25 is a one-line change if needed.
• GDP-free impulse. The canonical China credit impulse is new credit flow as a share of nominal GDP. With no GDP series in the pull, this uses the acceleration of TSF YoY growth as a monotone proxy; add quarterly nominal GDP to recover the % -of-GDP version.
• Managed FX. The RMB moves in policy-guided steps, so its information content is low — don't over-read FX contributions (hence the 0.10 weight).
• Property is lagged, monthly & partly administered — don't over-read week-to-week housing moves.
• Mixed frequency. Property + credit (50% of weight) are monthly; the index's week-to-week movement is driven by the daily half (FX / equity / rates), with the macro half stepping once a month.
• Structural breaks (the 2018 trade war, COVID, the 2021–22 property crisis, the 2023–24 deflation scare) distort any fixed base and short rolling window; inspect component z-scores around these dates.

10

References

Goldman Sachs Global Economics. Our New G10 Financial Conditions Indices, Global Economics Analyst, 20 April 2017. gspublishing.com

Bank for International Settlements. Effective exchange rate indices (REER, broad) and residential property price statistics.

People's Bank of China. Total Social Financing aggregates and SHIBOR; National Bureau of Statistics — 70-city residential price indices; ChinaBond — government-bond yield curve.